← All posts

Why does your scan keep coming out crooked?

The scan is readable. It is just leaning, by two or three degrees, the way a photocopy from 1994 leans. You did not put the paper in crooked — you squared the stack against the guide and watched it go in.

And it is not every page. Page 4 is fine, page 5 leans left, page 6 leans left by a different amount. Which rules out the theory that you are bad at putting paper in trays.

Everything written about this tells you how to straighten it, and nothing tells you what the machine is doing — which is the part worth knowing, because a scanner that does this today will do it again on Thursday.

What the feeder is doing to your paper

A sheet-fed scanner does not photograph a page. It pulls the page past a sensor bar that reads one thin line at a time and builds the image out of those lines in the order they arrive. The paper moves; the camera does not. So how the scan looks is a statement about how the paper travelled.

The paper is driven by a rubber pick roller pressing it against a separation roller that pushes back, so only one sheet goes at a time. That pair grips the sheet across a narrow strip. If the grip is not even across that strip — more friction on the left than on the right — the sheet does not only move forward, it turns. A fraction of a degree per centimetre of travel is enough, and by the time the trailing edge is through, the page is three degrees out.

Which makes every cause worth checking a way for one side of a sheet to be gripped harder than the other:

The rollers are glazed, not worn out. Paper sheds a fine dust, and toner and paper coating polish the rubber into a hard shiny film, and a glazed patch slips. Rollers rarely glaze evenly: the middle sees every sheet and the ends see only the wide ones. That is why the fault comes on gradually, and why cleaning fixes it more often than replacing does.

The edge guides were left wide. Guides do not straighten paper; they stop it wandering. A stack sitting 5 mm clear of both guides enters the nip already rotated, and the nip faithfully preserves the angle it was given.

The stack was not jogged. If the leading edges are not flush, each sheet meets the rollers at its own angle — the fingerprint of a different angle on every page.

One sheet is not like the others. A dog-eared corner, a staple hole, a fold, a sheet that came out of a printer curled: each changes where the grip starts. One bad page in twenty is the paper.

Most feeders try to correct all of this by driving the sheet into a stalled roller pair so the leading edge buckles square before the drive starts. It works well on ordinary office paper and much less well on thin, stiff or glossy stock, which is why the same scanner handles your letters perfectly and makes a mess of a stack of invoices on card.

Read the angles, not the pages

You do not have to guess which of those it is. The pattern across the document names the cause.

Three things that look like skew and are not

A book on the glass. Near the spine the paper lifts away from the platen and curves, and the lines of type curve with it. Nothing is tilted; the page is warped, and no rotation undoes a curve. That needs dewarping, or a flatter original.

The sheet turned while it was moving. The image is built line by line as the paper travels, so if the sheet rotates mid-feed, the top of the page was captured at one angle and the bottom at another. That is a sheared page, not a rotated one. The test takes five seconds: lay a straight edge along the left margin of the text. Straight but sloped is honest skew and can be corrected. If the margin bends, no amount of turning fixes it, and re-feeding that one sheet is quicker than anything else you might try.

A photograph of a page. Hold a phone over a document and you get perspective: the near edge of the paper is longer than the far edge, so the page is a trapezium, not a tilted rectangle. Rotating a trapezium gives you a rotated trapezium. That needs a four-corner correction — the document mode in your phone's camera app — and no page-straightening tool can do it. A quarter turn is a different thing again and usually looks after itself: our JPG to PDF reads a JPEG's EXIF orientation tag, so a photo taken with the phone held sideways comes out the right way up in the PDF, and when one does not, the tag was stripped on the way.

Ten seconds more before you reach for any tool: if the crooked page has real, selectable text on it, nothing scanned it — someone drew the content at an angle. Our PDF inspector says whether a page carries a real text layer without uploading the file anywhere, and OCR PDF gives the same answer as a per-page list for a long document. Read its notice first: it reports which pages are text and which are pictures, and cannot add a text layer, because the recognition engine is not installed here.

Straightening the pages you already have

  1. Crop first if there is a border. A black band, a strip of the next sheet, punch holes down the margin — all of it is ink as far as an angle measurement is concerned, and a solid dark edge drags the answer. Crop PDF takes millimetres off any side.
  2. Set the page range to the pages that are actually crooked. Pages you leave out are copied across byte for byte, keeping their text, links and sharpness.
  3. Leave the tilt range at 10 degrees for prose, and drop it to 3 or 4 for forms and tables. The angle is found by squashing the page's ink onto the vertical axis and looking for the angle at which the ink piles into distinct rows — which is exactly what lines of type do, and exactly what a page of ruled boxes does not. A big search range on a table gives the search more wrong answers to find.
  4. Raise the resolution above the 200 dpi default if the scan was made at 300 or more. The corrected page is redrawn at whatever you set, and what it drops is not recoverable afterwards.
  5. Read the reported angles. Deskew PDF tells you what it found on each page — "page 7 was turned 2.4° clockwise" — and that list is the diagnostic from the section above, handed to you for free.

Stopping it at the machine, which is the only permanent fix

Five minutes here beats straightening every batch for the next year.

When a desktop tool is better

If this is a weekly job, or the document is a book, stop doing it in a browser.

What straightening will not do

It cannot tell you which of the causes above is yours. It measures the file. The angles are evidence; the diagnosis is still a person looking at a machine.

A corrected page stops being text. The page has to be resampled to be turned, so it comes back as a picture: no selecting, no searching, no screen reader, no links or comments. Pages it decides were already level keep everything, which is why the page range matters.

The file gets bigger. Every corrected page becomes a JPEG. Run Compress PDF afterwards if it has to be emailed — a page of grey pixels is exactly the case compression is good at, so the cost is usually recovered.

Under a tenth of a degree, it does nothing at all and says so. If it reports that every page was already level and you can still see a slope, believe the tool: what you are looking at is a curve, a shear or a perspective, and the answer is in the section above rather than in a straightening tool.